[llvm] Add optimisation for SVE intrinsics with no active lanes (PR #73964)

[Maciej Gabka via llvm-commits]

================
@@ -1406,9 +1406,30 @@ static std::optional<Instruction *> instCombineSVEAllActive(IntrinsicInst &II,
   return &II;
 }
 
+// Optimize operations that take an all false predicate or send them for
----------------
mgabka wrote:

Hi @MarkAHarley,
 This comment is misleading, as it is missing the point that canonicalization applied here is only for the ptrue all predicate.
I would suggest rephrase it.

"Simplify operations where predicate has all inactive lanes or try to replace with _u form when all lanes are active".

 

https://github.com/llvm/llvm-project/pull/73964